Equipment monitoring prediction method and system based on industrial middle station and edge computing

Abstract

The invention discloses an equipment monitoring prediction method and system based on industrial middle platform and edge computing, and the system employs a monitoring collection unit to collect the vibration, temperature and current data of equipment in real time, and carries out the analysis through an innovative three-layer prediction architecture: deploying a lightweight threshold rule model at an edge layer to carry out the real-time anomaly detection, an XGBoost model is deployed in the middle layer to execute trend analysis, a time convolution network model is deployed in the cloud layer to realize deep time sequence analysis, and a corresponding grading early warning mechanism is established. The technical problems that an early warning mechanism is not intelligent enough, a prediction model is single, data preprocessing is not perfect enough and the like in traditional equipment monitoring are solved. Through cooperation of three layers of prediction architectures, omnibearing monitoring is realized, system response delay is reduced, and early warning accuracy is improved. Through the combination of a grading early warning mechanism and parameter correlation analysis, the system reliability is enhanced, and the equipment operation and maintenance efficiency is improved; and an edge calculation mode is adopted, so that the network bandwidth requirement is effectively reduced.

Description

Technical Field

[0001] The present invention relates to the field of industrial equipment monitoring technology, and in particular to an equipment monitoring prediction method and system based on industrial middleware and edge computing. Background Art

[0002] With the rapid development of Industry 4.0 and intelligent manufacturing, equipment health management and predictive maintenance have become key links to ensure the stable operation of industrial production. Unplanned downtime of industrial equipment causes huge economic losses to the manufacturing industry every year, and the use of predictive maintenance can significantly reduce equipment downtime and greatly improve the comprehensive utilization rate of equipment. The monitoring and prediction system of industrial equipment can effectively prevent equipment failures, reduce unplanned downtime, and improve equipment utilization by collecting equipment operation data in real time, combining artificial intelligence and big data analysis technology. At present, there are many technical solutions in the field of industrial equipment monitoring and prediction, but there are still many challenges in practical applications:

[0003] 1) From the perspective of the prediction model architecture, traditional single-model prediction solutions often use a unified algorithm framework to handle all prediction tasks, which makes it difficult to balance the needs of real-time performance and prediction accuracy. For example, motor bearing fault warning requires extremely fast response speed, while equipment life prediction requires in-depth analysis of comprehensive historical data. It is difficult to meet both types of requirements using the same prediction model, resulting in inefficient system resource utilization and unsatisfactory prediction results.

[0004] 2) From the perspective of system deployment, although pure cloud solutions have powerful computing capabilities, they have serious system response delay problems. Take large steel companies as an example. Their rolling mill equipment continuously generates a large amount of real-time data. If all of it is uploaded to the cloud for processing, it will cause a large network delay. This delay not only puts huge pressure on network bandwidth, but may also affect the stability of the monitoring system due to network fluctuations, and cannot meet the strict real-time requirements of industrial sites.

[0005] 3) From the perspective of monitoring mechanism, the existing simple threshold monitoring scheme lacks the ability to analyze the correlation between multiple parameters of equipment. For example, in industrial pump systems, parameters such as vibration, temperature, and current often have complex coupling relationships. Independent monitoring of a single parameter cannot detect complex fault modes such as abnormal vibration caused by bearing overheating, which can easily cause false alarms or missed alarms, affecting the accuracy of early warnings. Multi-parameter correlation analysis can significantly improve the accuracy of fault early warnings.

[0006] 4) From the perspective of early warning mechanism, existing technologies generally lack a systematic hierarchical early warning design. In actual industrial scenarios, equipment anomalies can be divided into different levels such as early warning, performance degradation, and fault warning, and different maintenance strategies are required for each level. However, the existing binary judgment mechanism (normal / abnormal) cannot provide such fine-grained early warning information, resulting in unreasonable allocation of maintenance resources. For example, for a slight performance degradation, it may only need to be handled at the next planned maintenance, while a serious fault requires immediate shutdown for maintenance.

[0007] In addition, with the development of edge computing technology, its application in industrial scenarios is becoming more mature. Edge computing can significantly reduce data processing delays and drastically reduce bandwidth usage.

[0008] Therefore, there is an urgent need for an equipment monitoring and prediction system that can comprehensively utilize the advantages of edge computing and industrial middleware, and has multi-level prediction capabilities and intelligent early warning mechanisms. The system should be able to achieve the coordinated cooperation of the three-layer prediction architecture, the correlation analysis of multiple parameters, and the intelligent classification of the early warning mechanism, so as to comprehensively improve the monitoring and prediction effect and operation and maintenance efficiency of industrial equipment. Summary of the invention

[0009] The problem to be solved by the present invention is how to provide an equipment monitoring and prediction method and system based on industrial middleware and edge computing to solve the technical problems existing in traditional equipment monitoring and prediction, such as large response delay, single early warning mechanism, and insufficient parameter correlation analysis.

[0010] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0011] An equipment monitoring and prediction method based on industrial middle platform and edge computing is characterized by comprising the following steps: S1. collecting operation data including vibration data, temperature data and current data, setting the sampling timing according to the equipment operation characteristics, collecting data according to the preset sampling frequency and preliminarily storing it in the edge controller;

[0012] S2. Distribute the operation data to the three-layer prediction model after preprocessing, wherein the preprocessing includes data cleaning, data repair and exception processing of the operation data, and the three-layer prediction model includes a lightweight threshold rule model deployed at the edge layer, an XGBoost model deployed at the middle platform layer, and a time convolutional network model deployed at the cloud layer; wherein the middle platform layer is an industrial middle platform;

[0013] S3. Use the three-layer prediction model to perform equipment status analysis and prediction, wherein the lightweight threshold rule model of the edge layer performs anomaly detection based on the operating data and the parameter reference interval, the XGBoost model of the middle platform layer performs feature analysis and trend prediction on the operating data, and performs anomaly judgment based on the parameter reference interval, and the time convolution network model of the cloud layer performs deep time series analysis and long-term prediction on the operating data, and performs risk assessment based on the parameter reference interval;

[0014] The parameter reference interval includes a vibration parameter reference interval, a temperature parameter reference interval and a current parameter reference interval. The vibration parameter reference interval is a range of a mean value plus or minus a standard deviation calculated based on the vibration data. The temperature parameter reference interval is a range of a mean value plus or minus a standard deviation calculated based on the temperature data. The current parameter reference interval is a range of a mean value plus or minus a standard deviation calculated based on the current data.

[0015] S4. A hierarchical warning mechanism is established based on the prediction results of the three-layer prediction model. When the edge layer detects an abnormality in the parameter reference interval, a first-level warning is triggered. When the middle platform layer detects an abnormal trend in the operation data, a second-level warning is triggered. When the cloud layer predicts an operation risk, a third-level warning is triggered.

[0016] S5. Respond and process according to the warning results of the hierarchical warning mechanism. When the first-level warning is triggered, the equipment stop operation instruction is sent to the equipment control system and the emergency warning information is pushed to the operation and maintenance personnel. When the second-level warning is triggered, the equipment inspection task is generated and the equipment maintenance plan is adjusted. When the third-level warning is triggered, the cloud layer generates an equipment operation and maintenance analysis report based on the predicted data;

[0017] Furthermore, the lightweight threshold rule model of the S3 edge layer performs parameter reference interval calculation and anomaly detection based on the operation data, specifically including:

[0018] Dynamically monitoring the operation data using a sliding window based on a time series, wherein the sliding window moves on the time axis according to a preset step length and calculates the mean and standard deviation within the sliding window, and when it is detected that a parameter exceeds the parameter reference interval, performing parameter association analysis;

[0019] The parameter association analysis includes counting the frequency of the vibration data, the temperature data and the current data exceeding the parameter reference interval in the sliding window, generating a parameter abnormality association matrix, and when a certain parameter exceeds the parameter reference interval, checking whether other corresponding parameters in the parameter abnormality association matrix are abnormal at the same time, and triggering the early warning mechanism in S4 if abnormality is confirmed;

[0020] During a time period when the parameter abnormality association matrix has not confirmed an abnormality, the statistical value of the sliding window is regularly added to the historical data to recalculate the parameter reference interval.

[0021] Furthermore, the XGBoost model of the S3 middle platform layer performs feature analysis and trend prediction on the operation data, specifically including:

[0022] Constructing characteristic matrices for the vibration data, the temperature data, and the current data in the operation data, respectively, wherein Fourier transform is performed on the vibration data to obtain a frequency spectrum distribution matrix of the vibration data, the mean and standard deviation of the temperature data are calculated to construct a statistical characteristic matrix of the temperature data, and the effective value of the current data is calculated to construct a load characteristic matrix of the current data;

[0023] A comprehensive feature matrix of the operation data is formed by aligning the timestamps of the spectrum distribution matrix, the statistical feature matrix, and the load feature matrix, and the XGBoost model determines the splitting gain of each dimensional feature in the comprehensive feature matrix through iterative calculation to generate a feature weight coefficient;

[0024] The XGBoost model generates a device state prediction value based on the comprehensive feature matrix and the feature weight coefficient, and when the device state prediction value exceeds the parameter reference interval and meets the preset accuracy condition, the early warning mechanism in S4 is triggered;

[0025] The XGBoost model sets a performance evaluation cycle, and calculates the prediction accuracy during the evaluation cycle. When the prediction accuracy is lower than a preset threshold, incremental training is performed using the newly added operating data to update the feature weight coefficient.

[0026] Furthermore, the S3 cloud-layer temporal convolutional network model performs deep time series analysis and long-term prediction on the operation data, specifically including:

[0027] constructing the vibration data, the temperature data and the current data in the operation data into a time series data matrix, and extracting training samples from the time series data matrix according to a preset sampling interval;

[0028] The temporal convolutional network model includes a causal convolution structure and a dilated convolution structure, wherein the causal convolution structure is used to extract temporal features, and the dilated convolution structure is used to expand the receptive field range; the operating state feature matrix is ​​constructed for the training samples through the causal convolution structure, and the operating state feature matrix is ​​mapped to the deep feature space using the dilated convolution structure to form a state representation vector of the operating data;

[0029] The temporal convolutional network model generates a predicted value of the operating data based on the state representation vector of the operating data, and when the predicted value continuously deviates from the parameter reference interval within a preset time range, the early warning mechanism in S4 is triggered;

[0030] The temporal convolutional network model sets a model update trigger condition, wherein when the number of the newly added training samples reaches a preset sample threshold, the temporal convolutional network model is retrained using the historical data in the time series data matrix and the newly added training samples, and the state representation vector is updated.

[0031] The present invention also discloses an equipment monitoring and prediction system based on industrial middle platform and edge computing, which is characterized by comprising: a monitoring and acquisition unit, an edge computing unit, an industrial middle platform unit and a cloud processing unit; wherein the monitoring and acquisition unit comprises a sensor component and a data acquisition component, and the sensor component comprises a vibration sensor, a temperature sensor and a current sensor, wherein the vibration sensor is arranged on the main bearing housing of the equipment, the temperature sensor is arranged in the heating area of ​​the equipment, and the current sensor is arranged on the power supply line of the equipment, and the data acquisition component is electrically connected to the sensor component, and is used to collect equipment operation data according to a preset sampling frequency;

[0032] The edge computing unit is connected to the monitoring and acquisition unit through a field bus, and is provided with a data preprocessing module and an edge analysis module. The data preprocessing module is used to clean and repair the equipment operation data. The edge analysis module has a built-in lightweight threshold rule model for real-time anomaly detection, and sends an equipment stop operation instruction to the equipment control system and pushes emergency alarm information to the operation and maintenance personnel when the first-level warning is triggered;

[0033] The industrial middle platform unit is data-connected with the edge computing unit, and is provided with a data analysis module, a trend prediction module and a task management module. The data analysis module is used to construct a feature matrix and perform feature analysis of the middle platform layer. The trend prediction module has a built-in XGBoost prediction model for performing trend prediction analysis of the middle platform layer. The task management module is used to generate equipment inspection tasks and adjust maintenance plans when a secondary warning is triggered;

[0034] The cloud processing unit is data-connected with the edge computing unit and the industrial middle platform unit, and is provided with a deep learning module and an early warning processing module. The deep learning module has a built-in time convolution network model for long-term prediction. The early warning processing module is used to generate an equipment operation and maintenance analysis report based on the prediction data when the third-level early warning is triggered;

[0035] The beneficial effects of the present invention are as follows:

[0036] (1) The present invention adopts an innovative combination of three-layer prediction architectures. Through the coordinated cooperation of the lightweight threshold rule model at the edge layer, the XGBoost model at the middle platform layer, and the temporal convolutional network model at the cloud layer, it achieves full coverage from real-time monitoring to long-term prediction, effectively reduces system response delay, and improves prediction accuracy.

[0037] (2) The present invention designs a systematic hierarchical warning mechanism, which closely combines the first-level warning, the second-level warning and the third-level warning with the three-layer architecture to form a complete warning response system, realizing comprehensive warning from emergency exception handling to long-term risk prediction.

[0038] (3) The present invention uses a sliding window for dynamic monitoring and constructs a parameter anomaly association matrix, thereby realizing cross-validation of multi-dimensional parameters such as vibration, temperature, and current, and improving the accuracy and reliability of anomaly detection.

[0039] (4) The present invention selects the most suitable algorithm model at each level. The lightweight rule model of the edge layer meets the real-time requirements, the XGBoost model of the middle platform layer is suitable for medium-scale data analysis, and the time convolution network of the cloud layer is suitable for deep time series analysis, thereby realizing the reasonable allocation of computing resources and the optimization of system performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 This is a flow chart of the equipment monitoring prediction method based on industrial middleware and edge computing.

[0041] Figure 2 This is a three-layer prediction model architecture diagram.

[0042] Figure 3 This is the architecture diagram of the equipment monitoring and prediction system based on industrial middleware and edge computing. DETAILED DESCRIPTION

[0043] like Figure 1 As shown, an embodiment of the present invention discloses a device monitoring and prediction method based on industrial middle platform and edge computing. The method achieves full coverage from real-time monitoring to long-term prediction through a three-layer prediction architecture. The present invention deploys a lightweight threshold rule model at the edge layer for real-time anomaly detection, deploys an XGBoost model at the middle platform layer for trend prediction, and deploys a time convolution network model at the cloud layer for deep time series analysis, forming a complete prediction and analysis system. Combined with a hierarchical early warning mechanism, the present invention can detect equipment anomalies in a timely manner and perform corresponding level processing. The specific implementation steps are as follows:

[0044] S1: Data Collection

[0045] In this embodiment, the monitoring and acquisition unit includes a sensor component and a data acquisition component. The sensor component includes a vibration sensor, a temperature sensor and a current sensor, which are used to collect operation data including vibration data, temperature data and current data. The vibration sensor is arranged on the main bearing housing of the equipment, the temperature sensor is arranged in the heating area of ​​the equipment, and the current sensor is arranged on the power supply line of the equipment.

[0046] The installation positions of the sensors are reasonably arranged according to the operating characteristics of the equipment to ensure accurate operating data. During the data collection process, the system sets the corresponding sampling timing according to the characteristics of different types of sensors to ensure the continuity and integrity of the data. Preferably, the sampling frequency of the vibration sensor is set to 1kHz, the sampling frequency of the temperature sensor is set to 1Hz, and the sampling frequency of the current sensor is set to 10Hz. The data acquisition component is electrically connected to the sensor component, collects data according to the preset sampling frequency and performs preliminary storage in the edge controller.

[0047] S2: Data preprocessing and distribution

[0048] In this embodiment, the collected operation data is preprocessed and then distributed to the three-layer prediction model. Data preprocessing includes data cleaning, data repair and exception processing of the operation data. The three-layer prediction model includes a lightweight threshold rule model deployed at the edge layer, an XGBoost model deployed at the middle platform layer, and a time convolutional network model deployed at the cloud layer.

[0049] Specifically, the data cleaning process first performs outlier detection on the collected vibration data, temperature data, and current data, and removes data points that are obviously deviated from the normal range. Preferably, the 3σ criterion is used for outlier judgment, that is, data that exceeds the range of plus or minus three times the standard deviation of the mean value is marked as an outlier. For missing data points, the system uses a data repair method for processing. Preferably, a linear interpolation method is used to repair short-term data missing, and the statistical characteristic values ​​of historical data are used to fill in longer-term data missing.

[0050] The preprocessed data is distributed to the three-layer prediction model according to different time scales and computing requirements. Among them, the edge layer receives real-time data streams for rapid anomaly detection; the middle platform layer receives downsampled medium time scale data for trend analysis; the cloud layer receives long time series data for deep time series analysis. Preferably, the original sampled data is processed as follows: the data collected by the vibration sensor at a sampling frequency of 1kHz is converted into 1Hz data by calculating a mean for every 1000 data points in the middle platform layer; the data collected by the current sensor at a sampling frequency of 10Hz is converted into 1Hz data by calculating a mean for every 10 data points in the middle platform layer; the data collected by the temperature sensor at a sampling frequency of 1Hz remains unchanged. Through this downsampling process, the middle platform layer obtains unified 1Hz medium time scale data for trend analysis. Preferably, for the long time series data of the cloud layer, the system further downsamples the above 1Hz data: a set of statistical values, including mean, standard deviation, maximum value, and minimum value, are calculated for every 3600 data points (i.e., 1 hour of data), so as to obtain long time series data in hours, which is convenient for long-term trend analysis. Through this preprocessing and distribution mechanism, the present invention achieves data quality assurance and reasonable allocation of computing resources, providing a reliable data foundation for subsequent multi-level prediction analysis.

[0051] S3: Three-layer prediction model analysis and prediction

[0052] Reference Figure 2 In this embodiment, the system uses a three-layer prediction model to analyze and predict the device status. Specifically, different prediction models are deployed at the edge layer, the middle platform layer, and the cloud layer to achieve multi-level status monitoring and prediction analysis.

[0053] The lightweight threshold rule model of the edge layer performs anomaly detection based on the operating data and the parameter reference interval. Among them, the parameter reference interval includes a vibration parameter reference interval, a temperature parameter reference interval and a current parameter reference interval. The vibration parameter reference interval is the range of the mean plus or minus the standard deviation calculated based on the vibration data, the temperature parameter reference interval is the range of the mean plus or minus the standard deviation calculated based on the temperature data, and the current parameter reference interval is the range of the mean plus or minus the standard deviation calculated based on the current data. Preferably, the parameter reference interval uses the ±1σ (one standard deviation) range as the normal operating range, for example: when the vibration mean of a certain device is 2.5mm / s and the standard deviation is 0.3mm / s, the vibration parameter reference interval is [2.2mm / s, 2.8mm / s]; when the temperature mean is 45℃ and the standard deviation is 2℃, the temperature parameter reference interval is [43℃, 47℃]; when the current mean is 80A and the standard deviation is 5A, the current parameter reference interval is [75A, 85A].

[0054] Specifically, the edge layer uses a sliding window based on a time series to dynamically monitor the operation data. The sliding window moves on the time axis according to a preset step size and calculates the mean and standard deviation within the sliding window. When it is detected that the parameter exceeds the parameter reference interval, a parameter association analysis is performed. The parameter association analysis includes counting the frequency of the vibration data, the temperature data and the current data exceeding the parameter reference interval in the sliding window to generate a parameter abnormality association matrix. When a certain parameter exceeds the parameter reference interval, check whether other corresponding parameters in the parameter abnormality association matrix are abnormal at the same time, and if the abnormality is confirmed, trigger the early warning mechanism. Preferably, taking the operation process of an industrial pump as an example: when the sliding window detects that the bearing vibration value reaches 2.9mm / s and exceeds the reference interval [2.2mm / s, 2.8mm / s], the system finds through the parameter abnormality association matrix that the bearing temperature simultaneously rises to 48°C and exceeds the reference interval [43°C, 47°C], and the motor current rises to 87A and exceeds the reference interval [75A, 85A]. Since the three parameters are abnormal at the same time and are related (bearing failure is usually manifested as the coupling effect of increased vibration, temperature rise and current rise), the system confirms it as a real abnormality and triggers the early warning mechanism. In contrast, when only a short-term abnormality in the vibration value is detected while the temperature and current remain within the normal range, the system determines that it may be a temporary disturbance and does not immediately trigger the early warning mechanism.

[0055] Preferably, the time length of the sliding window is 10 minutes, and the preset step size is 1 minute, which ensures the balance between the continuity of data sampling and the computational efficiency. In each sliding window, the system records the maximum value, minimum value, mean value and standard deviation of each parameter to evaluate the stability of parameter changes. Preferably, the parameter abnormality association matrix adopts a three-dimensional matrix structure, in which the first dimension represents the parameter type (vibration, temperature, current), the second dimension represents the time window sequence number, and the third dimension represents the abnormal state mark (0 represents normal, 1 represents abnormal). Preferably, the parameter abnormality association matrix adopts a three-dimensional matrix structure with a dimension of 3×N×2, in which: the first dimension size is 3, corresponding to the three parameters of vibration, temperature, and current respectively; the second dimension size is N, representing the last N time windows, for example, N=10 represents the data of the last 10 time windows; the third dimension size is 2, which records whether the parameter value exceeds the reference interval (0 represents normal, 1 represents abnormal). For example, the matrix element [0][5][1] = 1 indicates that the abnormal flag bit (third dimension index 1) value of the vibration parameter (first dimension index 0) in the 6th time window (second dimension index 5) is 1, that is, an abnormality occurs; similarly, [1][5][1] = 1 indicates that the abnormal flag bit (third dimension index 1) value of the temperature parameter (first dimension index 1) in the 6th time window (second dimension index 5) is 1, that is, an abnormality occurs. When the system detects that [0][5][1] and [1][5][1] are 1 at the same time, it means that both the vibration and temperature parameters are abnormal in the same time window. In this case, the system can quickly confirm the correlation between the parameter abnormalities.

[0056] During the time period when the parameter anomaly association matrix has not confirmed anomalies, the statistical values ​​of the sliding window are regularly added to the historical data to recalculate the parameter reference interval. Preferably, the system performs an update calculation of the parameter reference interval every 24 hours, using the historical data of the last 7 days as the basis for calculation to ensure the timeliness and stability of the reference interval. Preferably, when performing parameter association analysis, the system sets the minimum anomaly duration threshold to 30 seconds, that is, only when the parameter anomaly duration exceeds this threshold will it be recorded in the parameter anomaly association matrix to filter out false alarms caused by instantaneous fluctuations.

[0057] At the middle platform layer, the XGBoost model performs feature analysis and trend prediction on the operating data. Specifically, first, feature matrices are constructed for the vibration data, the temperature data, and the current data in the operating data. The vibration data is Fourier transformed to obtain the spectrum distribution matrix of the vibration data, the mean and standard deviation of the temperature data are calculated to construct the statistical feature matrix of the temperature data, and the effective value of the current data is calculated to construct the load feature matrix of the current data. Preferably, the spectrum distribution matrix of the vibration data contains the amplitude of the main frequency components, and the matrix dimension is M×F, where M is the number of time windows and F is the number of frequency components. The statistical feature matrix of the temperature data has a dimension of M×2, including two statistics of the mean and standard deviation of each time window. The load feature matrix of the current data has a dimension of M×1, which records the effective value in each time window. Through this feature extraction method, the system can capture the frequency characteristics of vibration, the statistical characteristics of temperature, and the load characteristics of current respectively.

[0058] The comprehensive feature matrix of the operation data is formed by aligning the timestamps of the spectrum distribution matrix, the statistical feature matrix and the load feature matrix, and the XGBoost model determines the splitting gain of each dimensional feature in the comprehensive feature matrix through iterative calculation to generate feature weight coefficients. Preferably, the time alignment of the comprehensive feature matrix uses 1 minute as the basic time unit, downsampling is performed for data with a sampling frequency higher than 1 minute, and linear interpolation is used to supplement data with a sampling frequency lower than 1 minute to ensure the time consistency of the feature data.

[0059] The XGBoost model generates a predicted value of the equipment state based on the comprehensive feature matrix and the feature weight coefficient. When the predicted value of the equipment state exceeds the parameter reference interval and meets the preset accuracy condition, the early warning mechanism is triggered. Preferably, taking the prediction of fan vibration parameters as an example: the XGBoost model obtains a predicted value of 2.9mm / s based on the comprehensive feature matrix analysis, which exceeds the vibration parameter reference interval [2.2mm / s, 2.8mm / s]; at the same time, the model evaluates the accuracy of the prediction result to 92%, which is higher than the preset 85% accuracy threshold. At this time, the system triggers a secondary early warning. If the prediction accuracy is lower than the preset threshold, the system will not trigger an early warning even if the prediction value exceeds the parameter reference interval to avoid false alarms.

[0060] Preferably, the main parameters of the XGBoost model are configured as follows: the maximum tree depth is set to 6, the number of weak learners is set to 100, the learning rate is set to 0.1, the minimum split loss threshold is set to 1, the L1 regularization term coefficient is set to 0.1, and the L2 regularization term coefficient is set to 0.2.

[0061] The XGBoost model sets a performance evaluation cycle, and the prediction accuracy is calculated during the evaluation cycle. When the prediction accuracy is lower than the preset threshold, the newly added operation data is used for incremental training to update the feature weight coefficient. Preferably, the performance evaluation cycle is set to once every 8 hours, and the preset accuracy threshold is set to 85%. When performing incremental training, the historical data of the last 30 days is used as the training set, and the latest 1000 samples are retained for model verification each time. During the incremental training process, the learning rate is dynamically adjusted to 0.5 times the basic learning rate to avoid overfitting of the model. Preferably, in order to improve the reliability of the prediction, the system sets a confidence evaluation mechanism for the prediction results. Specifically, by calculating the variance of the predicted value, the prediction results are divided into three confidence levels: high, medium, and low, corresponding to prediction results with confidence greater than 90%, 70%-90%, and less than 70%, respectively. When the confidence of the prediction is lower than 70%, the system will mark a low confidence prompt in the warning information.

[0062] At the cloud layer, the time convolutional network model performs in-depth time series analysis and long-term prediction on the operating data. Specifically, the vibration data, the temperature data and the current data in the operating data are constructed into a time series data matrix, and training samples are extracted from the time series data matrix according to a preset sampling interval. Preferably, the construction of the time series data matrix adopts a sliding window method, the window length is set to 168 hours (7 days), and the step length is set to 1 hour to ensure the capture of the periodic characteristics of the equipment operation. For different types of data, time downsampling is performed according to their sampling frequency and uniformly converted into hourly data. Preferably, a stratified sampling strategy is adopted for the extraction of training samples, and the data is divided according to the ratio of 80% training set and 20% validation set. During the sample extraction process, it is ensured that samples of various operating conditions are included in the training set to improve the generalization ability of the model.

[0063] The temporal convolutional network model includes a causal convolution structure and a dilated convolution structure, wherein the causal convolution structure is used to extract temporal features, and the dilated convolution structure is used to expand the receptive field range. The operating state feature matrix is ​​constructed for the training sample through the causal convolution structure, and the operating state feature matrix is ​​mapped to the deep feature space by using the dilated convolution structure to form the state representation vector of the operating data. Preferably, the specific configuration of the causal convolution structure is as follows: the first layer convolution kernel size is 3, and the number of output channels is 32; the second layer convolution kernel size is 3, and the number of output channels is 64; the third layer convolution kernel size is 3, and the number of output channels is 128, and each layer of convolution is followed by a batch normalization layer and a ReLU activation function to avoid the gradient vanishing problem. Preferably, the dilated convolution structure adopts an increasing dilation rate setting: the first layer dilation rate is 1, and the receptive field is 3; the second layer dilation rate is 2, and the receptive field is 7; the third layer dilation rate is 4, and the receptive field is 15. By setting the dilation rate in an increasing manner, the model can effectively capture dependencies of different time scales.

[0064] The temporal convolutional network model generates a predicted value of the operating data based on the state characterization vector of the operating data. When the predicted value continuously deviates from the parameter reference interval within a preset time range, an early warning mechanism is triggered. Preferably, the generation of the predicted value adopts a uniform time interval to predict the equipment status for the next 7 days, and a prediction point is generated every 6 hours to evaluate the long-term operation risk of the equipment. When the predicted value continuously deviates from the parameter reference interval at multiple consecutive time points, a third-level early warning is triggered.

[0065] The temporal convolutional network model sets a model update trigger condition, wherein when the number of newly added training samples reaches a preset sample threshold, the temporal convolutional network model is retrained using the historical data in the time series data matrix and the newly added training samples, and the state representation vector is updated. Preferably, the trigger conditions for model update are as follows: the number of newly added samples reaches 20% of the original training set, the proportion of samples with prediction errors exceeding the preset threshold exceeds 15%, and the prediction accuracy for 7 consecutive days is less than 80%. Model update is triggered when any of the above conditions are met. Preferably, the model update process adopts the following strategy: retain the historical data of the last 90 days as the basic training set; use the transfer learning method to only fine-tune the parameters of the last two layers of the network; use an early stopping strategy during training, and stop training when the validation set loss has not improved for 5 consecutive epochs; the updated model needs to be verified by backtesting, and the accuracy rate is higher than the original model before it will be deployed online.

[0066] S4: Establishment of a graded early warning mechanism

[0067] In this embodiment, a hierarchical warning mechanism is established based on the prediction results of the three-layer prediction model, realizing all-round warning from real-time anomalies to long-term risks.

[0068] Specifically, when the edge layer detects an abnormality in the parameter reference interval, a first-level warning is triggered. The first-level warning is an emergency abnormality warning for the real-time operation status of the equipment. When the lightweight threshold rule model of the edge layer detects that the vibration data, temperature data or current data exceeds the respective parameter reference interval, and the abnormality is confirmed through the parameter abnormality association matrix, the system immediately triggers the first-level warning.

[0069] When the middle platform detects an abnormal trend in the operating data, a secondary warning is triggered. The secondary warning is a predictive warning for the equipment operation trend, which is triggered when the equipment status value predicted by the XGBoost model of the middle platform exceeds the parameter reference interval and meets the preset accuracy condition.

[0070] When the cloud layer predicts an operation risk, a third-level warning is triggered. The third-level warning is a predictive warning for the long-term operation risk of the equipment, which is triggered when the equipment state predicted by the temporal convolutional network model of the cloud layer continuously deviates from the parameter reference interval within a preset time range.

[0071] S5: Warning response processing

[0072] In this embodiment, the system performs response processing according to the warning results of the hierarchical warning mechanism, which is specifically achieved through the following technical methods:

[0073] When the first-level warning is triggered, the edge analysis module of the edge computing unit performs the following response processing: sending a device stop operation instruction to the device control system through the industrial communication protocol, and pushing emergency warning information to the operation and maintenance personnel. Preferably, the edge analysis module uses the Modbus TCP protocol to write a shutdown instruction to the device's programmable logic controller (PLC), and at the same time pushes the alarm information to the operation and maintenance personnel's mobile terminal in real time through WebSocket. The alarm information is encapsulated in JSON format and contains fields such as device ID, abnormal parameters, current parameter values, and trigger time.

[0074] When the secondary warning is triggered, the task management module of the industrial middle platform unit performs the following response processing: generating equipment inspection tasks and adjusting equipment maintenance plans. Preferably, the task management module creates inspection task records in the maintenance management database, including fields such as work order number, equipment number, inspection items, and planned time, and updates relevant records of the maintenance plan.

[0075] When the third-level warning is triggered, the warning processing module of the cloud processing unit performs the following response processing: generating an equipment operation and maintenance analysis report based on the predicted data. Preferably, the warning processing module uses the predicted data to generate equipment performance trend analysis, fault risk assessment and other content, and sends the analysis report to designated managers and technicians by email to achieve timely delivery of risk warning information.

[0076] like Figure 3 As shown, the present invention also discloses an equipment monitoring and prediction system based on industrial middle platform and edge computing, the system includes a monitoring and acquisition unit, an edge computing unit, an industrial middle platform unit and a cloud processing unit, and each unit realizes data interconnection through the industrial Ethernet. The system realizes all the functions of the aforementioned equipment monitoring and prediction method, and the specific implementation method of each unit is described in detail below.

[0077] In this embodiment, the monitoring and acquisition unit includes the sensor assembly and the data acquisition assembly. The sensor assembly includes the vibration sensor, the temperature sensor and the current sensor. As mentioned above, the vibration sensor is arranged on the main bearing housing of the equipment, the temperature sensor is arranged in the heating area of ​​the equipment, and the current sensor is arranged on the power supply line of the equipment. The data acquisition assembly is electrically connected to the sensor assembly and is used to collect the equipment operation data according to the preset sampling frequency. Preferably, the data acquisition assembly adopts a modular design, including the multi-channel analog signal acquisition card and the data preprocessing unit, and the data preprocessing unit is equipped with an industrial-grade storage card for local data caching.

[0078] The edge computing unit is communicatively connected with the monitoring and acquisition unit via the field bus, and is provided with the data preprocessing module and the edge analysis module. The data preprocessing module is used to clean and repair the equipment operation data, and the edge analysis module is equipped with the lightweight threshold rule model for real-time anomaly detection, and when the first-level warning is triggered, it sends the equipment stop operation instruction to the equipment control system and pushes the emergency warning information to the operation and maintenance personnel. Preferably, the edge computing unit adopts an industrial-grade edge computing gateway, and the edge analysis module realizes real-time push of warning information through the publish-subscribe mode.

[0079] The industrial middle platform unit is data-connected to the edge computing unit, and is provided with the data analysis module, the trend prediction module and the task management module. The data analysis module is used to construct the feature matrix and perform feature analysis of the middle platform layer. The trend prediction module has the XGBoost prediction model built in to perform trend prediction analysis of the middle platform layer. The task management module is used to generate the equipment inspection task and adjust the maintenance plan when the secondary warning is triggered. Preferably, the industrial middle platform unit adopts an enterprise-level server configuration, and accelerates feature extraction and model calculation through the parallel computing framework.

[0080] The cloud processing unit is data-connected with the edge computing unit and the industrial middle-end unit, and is provided with the deep learning module and the early warning processing module. The deep learning module has the temporal convolutional network model built in for the long-term prediction, and the early warning processing module is used to generate the equipment operation and maintenance analysis report based on the prediction data when the third-level early warning is triggered. Preferably, the cloud processing unit adopts the containerized deployment solution and is equipped with a deep learning acceleration unit for model training and reasoning. The message queue mechanism is used to realize data transmission between the various units of the system, and the encryption protocol is used to ensure the security of data transmission. The overall system is configured and managed through the Web management interface, supports centralized log collection and analysis, and realizes real-time monitoring of system performance.

[0081] Through the implementation of the above technical solutions, this system realizes all-round monitoring and predictive maintenance of industrial equipment. The various units of the system work together, the edge layer ensures real-time response, the middle platform layer realizes trend analysis, and the cloud layer completes in-depth prediction, forming a complete monitoring and prediction system, which effectively improves the efficiency of equipment operation and maintenance and reduces the risk of equipment failure. The layered architecture design of this system fully utilizes the advantages of edge computing, the industrial middle platform and the cloud computing, and realizes the reasonable allocation of computing resources and the optimization of data processing efficiency.

Claims

1. A device monitoring prediction method based on industrial middle platform and edge computing, characterized in that: The following steps are involved: S1. Collect operation data including vibration data, temperature data and current data, set the sampling sequence according to the operation characteristics of the equipment, collect data according to the preset sampling frequency and perform preliminary storage in the edge controller; S2. Distribute the operation data to the three-layer prediction model after preprocessing, wherein the preprocessing includes data cleaning, data repair and exception processing of the operation data, and the three-layer prediction model includes a lightweight threshold rule model deployed at the edge layer, an XGBoost model deployed at the middle platform layer, and a time convolutional network model deployed at the cloud layer; wherein the middle platform layer is an industrial middle platform; S3. Use the three-layer prediction model to perform equipment status analysis and prediction, wherein the lightweight threshold rule model of the edge layer performs anomaly detection based on the operating data and the parameter reference interval, the XGBoost model of the middle platform layer performs feature analysis and trend prediction on the operating data, and performs anomaly judgment based on the parameter reference interval, and the time convolution network model of the cloud layer performs deep time series analysis and long-term prediction on the operating data, and performs risk assessment based on the parameter reference interval; The parameter reference interval includes a vibration parameter reference interval, a temperature parameter reference interval and a current parameter reference interval. The vibration parameter reference interval is a range of a mean value plus or minus a standard deviation calculated based on the vibration data. The temperature parameter reference interval is a range of a mean value plus or minus a standard deviation calculated based on the temperature data. The current parameter reference interval is a range of a mean value plus or minus a standard deviation calculated based on the current data. S4. A hierarchical warning mechanism is established based on the prediction results of the three-layer prediction model. When the edge layer detects an abnormality in the parameter reference interval, a first-level warning is triggered. When the middle platform layer detects an abnormal trend in the operation data, a second-level warning is triggered. When the cloud layer predicts an operation risk, a third-level warning is triggered. S5. Respond and process according to the warning results of the hierarchical warning mechanism. When the first-level warning is triggered, a command to stop the equipment operation is sent to the equipment control system and an emergency warning message is pushed to the operation and maintenance personnel. When the second-level warning is triggered, an equipment inspection task is generated and the equipment maintenance plan is adjusted. When the third-level warning is triggered, the cloud layer generates an equipment operation and maintenance analysis report based on the predicted data.

2. According to claim 1, a device monitoring prediction method based on industrial middle platform and edge computing is characterized in that: The lightweight threshold rule model of the S3 edge layer performs parameter benchmark interval calculation and anomaly detection based on the operation data, specifically including: Dynamically monitoring the operation data using a sliding window based on a time series, wherein the sliding window moves on the time axis according to a preset step length and calculates the mean and standard deviation within the sliding window, and when it is detected that a parameter exceeds the parameter reference interval, performing parameter association analysis; The parameter association analysis includes counting the frequency of the vibration data, the temperature data and the current data exceeding the parameter reference interval in the sliding window, generating a parameter abnormality association matrix, and when a certain parameter exceeds the parameter reference interval, checking whether other corresponding parameters in the parameter abnormality association matrix are abnormal at the same time, and triggering the early warning mechanism in S4 if abnormality is confirmed; During a time period when the parameter abnormality association matrix has not confirmed an abnormality, the statistical value of the sliding window is regularly added to the historical data to recalculate the parameter reference interval.

3. According to the equipment monitoring prediction method based on industrial middle platform and edge computing according to claim 1, it is characterized in that: The XGBoost model of the S3 middle platform layer performs feature analysis and trend prediction on the operation data, specifically including: Constructing characteristic matrices for the vibration data, the temperature data, and the current data in the operation data, respectively, wherein Fourier transform is performed on the vibration data to obtain a frequency spectrum distribution matrix of the vibration data, the mean and standard deviation of the temperature data are calculated to construct a statistical characteristic matrix of the temperature data, and the effective value of the current data is calculated to construct a load characteristic matrix of the current data; A comprehensive feature matrix of the operation data is formed by aligning the timestamps of the spectrum distribution matrix, the statistical feature matrix, and the load feature matrix, and the XGBoost model determines the splitting gain of each dimensional feature in the comprehensive feature matrix through iterative calculation to generate a feature weight coefficient; The XGBoost model generates a device state prediction value based on the comprehensive feature matrix and the feature weight coefficient, and when the device state prediction value exceeds the parameter reference interval and meets the preset accuracy condition, the early warning mechanism in S4 is triggered; The XGBoost model sets a performance evaluation cycle, and calculates the prediction accuracy during the evaluation cycle. When the prediction accuracy is lower than a preset threshold, incremental training is performed using the newly added operating data to update the feature weight coefficient.

4. According to claim 1, a device monitoring prediction method based on industrial middle platform and edge computing is characterized in that: The S3 cloud-layer temporal convolutional network model performs deep time series analysis and long-term prediction on the operation data, specifically including: The vibration data, the temperature data and the current data in the operation data are constructed into a time series data matrix, and training samples are extracted from the time series data matrix according to a preset sampling interval; the temporal convolution network model includes a causal convolution structure and a dilated convolution structure, wherein the causal convolution structure is used to extract time series features, and the dilated convolution structure is used to expand the receptive field range; an operation state feature matrix is ​​constructed for the training samples through the causal convolution structure, and the operation state feature matrix is ​​mapped to a deep feature space using the dilated convolution structure to form a state representation vector of the operation data; The temporal convolutional network model generates a predicted value of the operating data based on the state representation vector of the operating data, and when the predicted value continuously deviates from the parameter reference interval within a preset time range, the early warning mechanism in S4 is triggered; The temporal convolutional network model sets a model update trigger condition, wherein when the number of the newly added training samples reaches a preset sample threshold, the temporal convolutional network model is retrained using the historical data in the time series data matrix and the newly added training samples, and the state representation vector is updated.

5. An equipment monitoring and prediction system based on industrial middle platform and edge computing, characterized in that: It includes monitoring and collection unit, edge computing unit, industrial middle platform unit and cloud processing unit; Wherein, the monitoring and acquisition unit includes a sensor component and a data acquisition component, the sensor component includes a vibration sensor, a temperature sensor and a current sensor, wherein the vibration sensor is arranged on the main bearing housing of the equipment, the temperature sensor is arranged in the heating area of ​​the equipment, the current sensor is arranged on the power supply line of the equipment, and the data acquisition component is electrically connected to the sensor component and is used to collect equipment operation data according to a preset sampling frequency; The edge computing unit is connected to the monitoring and acquisition unit through a field bus, and is provided with a data preprocessing module and an edge analysis module. The data preprocessing module is used to clean and repair the equipment operation data. The edge analysis module has a built-in lightweight threshold rule model for real-time anomaly detection, and sends an equipment stop operation instruction to the equipment control system and pushes emergency alarm information to the operation and maintenance personnel when the first-level warning is triggered; The industrial middle platform unit is data-connected with the edge computing unit, and is provided with a data analysis module, a trend prediction module and a task management module. The data analysis module is used to construct a feature matrix and perform feature analysis of the middle platform layer. The trend prediction module has a built-in XGBoost prediction model for performing trend prediction analysis of the middle platform layer. The task management module is used to generate equipment inspection tasks and adjust maintenance plans when a secondary warning is triggered; The cloud processing unit is data-connected to the edge computing unit and the industrial middle-end unit, and is provided with a deep learning module and an early warning processing module. The deep learning module has a built-in time convolutional network model for long-term prediction, and the early warning processing module is used to generate an equipment operation and maintenance analysis report based on the prediction data when the third-level early warning is triggered.

Images